スカイロジック List of Products and Services

We conducted a simple verification of dents and scratches on metal products such as motorcycle and automobile parts. Due to their almost identical appearance, there were incidents of shipping similar defective products, which caused issues. In the inspection process, automation is being promoted with the aim of pursuing stable inspection accuracy and cost reduction. As the shortage of personnel due to the aging workforce becomes more pronounced, we encourage you to consider improving operational efficiency through image inspection to protect "Made in Japan" manufacturing. 【Inspection Settings and Results】 Using the samples we have on hand for verification, we were able to detect dents and scratches. However, small dents and scratches present on good products, or surface conditions that appear as such in images, were detected as abnormalities. For extremely small items, we believe there are ways to address this by categorizing detection targets into "large dents" and "small dents," and only marking it as NG when a "large dent" is found.

Manufacturers dealing with various types of motors have shown interest in our AI-based image inspection software "DeepSky" since its launch. This time, it will assess scratches and dents on screws. Our inspection software has over 2,000 examples of image inspections. We have numerous inspection cases published, including metals, plastics, food, electronic circuit boards, and pharmaceuticals, so please check our website to see if there are similar cases to the inspections you are considering. 【Inspection Settings and Results】 As a result of verification using the samples we received, we were able to detect scratches, allowing us to report favorable inspection results. We can respond free of charge to requests for demonstrations, web meetings, and demo unit loan services. The loan period for demo units is approximately one week, but availability is limited, so there may be delays. The image on the left shows annotations, while the image on the right shows the detection frame with a favorable assessment.

This is a request for a simple verification from a manufacturer we have been working with for some time, who specializes in high-quality casting, machining, plastic processing, and the manufacturing and sales of acoustic products, delivering both domestically and internationally. This time, the focus is on cracks in the parts, and we attempted to analyze the provided images. Regarding "small cracks," we have been able to secure a sufficient number of samples, allowing us to hypothesize reasons for those that were not detected. However, for "large cracks" and "arm cracks," the limited number of samples resulted in insufficient findings in this verification. The report primarily features images of "small cracks," while other defects are included as reference images. Elements that are smaller, thinner, shorter, or darker than the cracks we were able to detect in this verification have not been identified. For "small, thin, and short" cracks, it is highly likely that they were not detected due to their size being too small relative to the overall field of view, and there is a possibility that dividing the field of view could allow for detection at the current resolution. For "thin and dark" cracks, it may be necessary to reconsider the imaging method. Additionally, if there are fundamentally difficult-to-detect types of defects, such as shallow cracks, the difficulty of detection increases significantly.

We identified the front and back of washers based on requests from manufacturers of high-performance automotive parts and precision machined products. Since we had several types of washers in stock, we conducted a simple inspection using a model we prepared instead of the samples sent by the customer. 【Inspection Setup and Results】 Upon checking with the washers we have, we found it possible to recognize those with and without a taper as different items, so we believe that the inspection you inquired about is likely feasible. In the images, we used our software called DeepSky to train the system, designating the side with the taper as "Front" and the flat side as "Back" for judgment. To conduct a more detailed verification, there are generally two methods available: (1) Providing images that can be inspected on the customer’s side: If you can provide images taken directly from above the inspection item with a fixed distance from the camera lens to the item, we can consider whether we can perform the inspection. (2) Using a loaned device for verification on the customer’s side: It is also possible for you to directly assess the inspection feasibility using our loaned equipment. Depending on the availability of the equipment, you may need to wait your turn.

This is an experiment to see if it is possible to distinguish a washer by annotating just one M5 washer in a single image and conducting 100 steps of learning with that one image. The left image shows the annotation (the task of enclosing to help remember what to find), while the right image places the M5 washer in the center with washers of different sizes on either side for verification. 【Inspection Settings and Results】 The washer could be found regardless of where it was placed on the screen. Even when multiple washers were placed and the lighting changed slightly, it was still able to identify them. The orientation did not matter, and for items with little variation, it could distinguish them with minimal learning. It recognized washers of different sizes and spring washers as washers. Nuts (with some shape variations) were also distinguished as washers. Items like mints, clips, and rollers in stock were identified as not being washers. It seems that items that are too large are not considered washers, while smaller items were recognized as washers. When an M5 washer was placed in between, it was clear that even though only one M5 washer had been taught, it was still able to make proper comparisons.

This is an inquiry from a manufacturer of special steel parts such as automotive components. We have decided to conduct a simple verification regarding cracks in the riveting area. In visual inspections conducted by workers, there is often variability in the criteria for determining "good" and "defective" products among inspectors, and the more ambiguous the judgment based on severity, the more inconsistent the inspection results tend to be. We propose stable and efficient inspections using image analysis. 【Inspection Settings and Results】 It was determined that inspecting cracks on the outer circumference is difficult, as they cannot be captured from above, and side imaging tends to misdetect shadows from the uneven shape. In the range we verified, diagonal surfaces cast shadows when imaged from above, making it impossible to capture the cracks themselves. Detection from the side is also unstable, making inspection of these diagonal surfaces challenging. However, cracks on the upper surface that can be captured from above were detectable. Additionally, the presence or absence of burrs was also detectable from above. Our company website offers a free trial of our inspection software online. You can actually experience inspection software that uses AI (deep learning).

This is support after the introduction of DeepSky, a manufacturer of sheet metal parts for transport machinery and construction machinery. DeepSky can be used for various parts and defects, but the detection accuracy may vary depending on the settings. Although it is a one-time purchase inspection software, we will continue to provide support after the introduction. This time, we received a consultation regarding difficulties with learning. 【Inspection Settings and Results】 We received information about the varieties that were not working well and removed the "OK" label from the annotations. We also adjusted the annotation frames to only cover the defective areas and labeled them with two types: "NG" and "NG Crack" for learning purposes. Since the annotation frames were relatively large, I suspect that during learning, the AI might have recognized it as "there is an NG item roughly in this position on the left" rather than "looking at the defect," which may have caused confusion when NG items appeared on the right side as well. The size of the annotation frames is a significant factor affecting the accuracy of the defect positions in the teacher images. 【Software Used】 Software used: DeepSky

We received samples from an aluminum processing manufacturer that we guided at the Sky Logic booth during the exhibition and conducted a simple evaluation. We attempted to detect "defects" and "scratches" on the workpiece. 【Inspection Settings and Results】 However, inspecting only one side does not allow for the inspection of all surfaces, so we proposed a method of moving the inspection item while changing the position of the lighting. We mounted a fixture on a rotary table for inspection. This time, we used three linear fluorescent lights as a way to highlight defects. 【Software and Equipment Used】 Software used: DeepSky Field of view: Approximately 220 x 175 mm Minimum size of inspection target: 2 mm Number of inspection points: 1 Camera resolution: 1.3 million pixels Lens focal length: 12 mm Distance between lens and product: Approximately 400 mm Lighting: Three linear fluorescent lights Distance between lighting and inspection item: One light was set to illuminate from a diagonal direction (approximately 30°) to ensure that light hits the sloped surface of the inspection item.

Our image inspection software is being considered in various fields and industries. The combination of AI (deep learning) and image inspection can be beneficial even in high-level solution areas. We recommend efficient inspections with DeepSky. [Inspection Settings and Results] The results of the verification conducted with the samples you provided showed that it was possible to detect "dents, impressions, chips, and large holes" and determine pass or fail. This time, we used a software called DeepSky, which utilizes AI (Deep Learning) for inspection. By training the software on the areas we want to detect, it can adjust its own setting parameters and recognize them. In actual operations, various patterns (sizes and features) of defects are expected, so it is necessary to increase the number of training images.

Metal products such as die-castings often reflect light, making inspection difficult in the past. With our inspection software DeepSky, released in 2020, it is now possible to operate even reflective metal products like die-castings with simple settings. 【Inspection Settings and Results】 By using DeepSky's inspection features, we were able to determine dents, impressions, and chip blockages in just 0.35 seconds. The image on the left shows the hole that should be present in a good product, while the image on the right is taken with the same color (white) as the good product's "hole," but it is accurately recognized as defective. The number at the top of the frame indicates the AI's confidence level (number of recognition points). 【Software and Equipment Used】 Software Used: DeepSky Learning Version Field of View: Approximately 56 x 42 mm Minimum Size of Inspection Target: 20 mm Number of Inspection Points: 1 overall Camera Resolution: 1.3 million pixels Lens Focal Length: 12 mm Distance Between Lens and Product: Approximately 150 mm Lighting: Flashlight (Chip blockage: spotlighting the processing hole)

We receive numerous inquiries about counting parts. Recently, we developed software for conveyor counting as an advanced application of DeepSky. We proposed a method where, while inspecting and counting, if you want to count 1,000 items, you can set the conveyor to light up at 990, have the workers line up in a row, and then stop the conveyor again at 1,000. In this inquiry, we attempted a test to count 23 correct workpieces while stationary on the workbench. 【Inspection Settings and Results】 In the settings shown in the left image, the parts to be counted are labeled as "Work," while other mixed parts are labeled as "Excluded." When set up as shown in the image above, the configuration passes if there are 20 counted parts (Work). The "Rs" on the right indicates the number of recognition points to be detected, set individually. The right image displays the detection screen. 【Software and Equipment Used】 Software Used: DeepSky Field of View: 280 x 200 mm Minimum Size of Inspection Target: 2 mm Number of Inspection Points: 8 Camera Resolution: 1.3 Megapixels Lens Focal Length: 6 mm Distance Between Lens and Product: Approximately 330 mm Lighting: Indoor Fluorescent Light

This is a request for a free evaluation to distinguish discoloration in parts (washers). We received an inquiry from a manufacturer that produces everything from fine electronic components to large fire-resistant glass. While looking forward to future developments, we conducted a simple verification test to identify the washers as an entry point. In the free sample evaluation, we will first conduct a simple verification using the samples or images provided and report the results. During the simple verification, we will evaluate whether the desired detection/judgment can be achieved using our internal equipment. If detection or judgment is possible in the simple verification, we recommend conducting a test (feasibility verification) assuming actual operation, where we will evaluate processing time and judgment accuracy. If feasibility verification is to be conducted, you can choose to continue with our company (for a fee) or use our loaned equipment for verification at your company. [Inspection Settings and Results] We labeled the normal washers as OK and the washers with part of them painted black as NG. We annotated six images with changed positions for the time being. The detection was accurate. It is possible to set how many OKs are needed to pass or if there is any NG to fail. Other detailed settings allow for recognition point counts and area judgments.

The industrial knife manufacturer focuses on high quality, precision, and functionality in the production of cutting tools for various objects, and is committed to pursuing product quality control. This time, we are verifying the judgment of defects in cutting machine parts. The current inspection method is conducted using visual checks and jigs, but there tends to be negligence during busy periods, leading to incidents that result in complaints. Since this is a common case with a variety of workpieces, we have also created a video to make it easier to visualize. Please take a look for reference. "Discovering the omission of two types of parts at the same time and stopping the conveyor" https://skylogiq.co.jp/DIY_HowTo/368 【Inspection Settings and Results】 It was possible to detect defective parts. However, due to the variety of workpieces, it is necessary to set an appropriate field of view for each size. There were also instances where defects could not be captured directly from above, necessitating angled imaging of holes, which is expected to make positioning and environmental conditions more stringent, as reported. 【Software Used】 Software used: DeepSky

Even manufacturers that globally deploy valve parts are considering our inspection software. Our software is utilized in specialized and intricate products across various industries. 【Inspection Settings and Results】 The task of enclosing the parts to be detected in a rectangle for parameter generation is called annotation. Each part is registered with the Label name shown in the right image, and it is set up so that if there is even one defect, it will be deemed a failure. We would like you to consider stable production through the automation of inspections, especially for products that require a high level of solution. 【Software and Equipment Used】 Software Used: DeepSky Learning Edition Field of View: Approximately 78 x 62 mm Minimum Size of Inspection Target: 2 mm Number of Inspection Points: 1 point, detecting defects from the entire screen Camera Resolution: 1.3 million pixels Lens Focal Length: 12 mm Distance Between Lens and Product: Approximately 150 mm Lighting: Ring lighting Distance Between Lighting and Inspection Item: Approximately 100 mm above

This is an external inspection at a metal parts manufacturer that produces chip saws. Although the inquiry was about dimensional angle inspection, upon discussion, it became clear that the request was for the detection of chipping and scratches. 【Inspection Setup and Results】 For the time being, we verified larger chipping and scratches, and it seemed that they could be detected without any issues. If it is something that can be detected by DeepSky (our AI inspection software), we can automatically conduct continuous inspections as shown in the video, stopping and notifying when a defect is found. We also conducted a demonstration where the saw's rotation was manually operated, and an NG was issued when an abnormality entered the field of view. 【Software Used】 Software Used: DeepSky Number of Inspection Points: 1 location, entire screen

The label equipment manufacturer has also been packaging needles and inquired about automating the visual inspection to ensure they are properly packaged. We will provide a free evaluation to determine whether we can identify "duplicates," "tilts," and "vertical misalignments" of the products after packaging. 【Inspection Settings and Results】 After conducting training with the samples you sent, we were able to identify good products, duplicates, tilts, and vertical misalignments. For inspection, it is necessary to transport the sheet as parallel as possible to the camera angle. We have set the field of view to evaluate 10 items in a single shot and inspection. The left image shows the display indicating that "10 good products" have been detected within the screen. The right image shows that "9 good products and 1 duplicate" have been detected, resulting in a failure judgment. 【Software and Equipment Used】 Software Used: DeepSky Learning Version Field of View: Approximately 78 x 62 mm Minimum Size of Inspection Target: 2 mm Number of Inspection Points: 1 (Pass if there are 10 good products within the screen) Camera Resolution: 1.3 million pixels Lens Focal Length: 25 mm Distance Between Lens and Product: Approximately 485 mm Lighting: Indoor fluorescent lights

This is an inquiry from a manufacturer of aluminum products and heat transfer processing with whom we have had a business relationship for some time. We will detect three types of defects: "something like dirt," "linear marks," and "dot-like marks." Sending actual samples will allow us to provide the most reliable report. In that case, we will need you to send several good products and workpieces with the defects you want to detect. Please contact us for more details. [Inspection Settings and Results] Since the scratches were of different types, we labeled them as "A / B / C" and registered them, making it possible to detect the scratches. The process of enclosing the areas we want to detect in a rectangle, as shown in the left image, is called annotation. We verified this using DeepSky, which can detect without fixed positioning, even at various orientations and angles during the work. DeepSky is software that excels in such inspections and is utilized across various industries.

We received a request for inspection of deformation of clips from a manufacturer specializing in renovations and remodeling. The inspection system was specifically envisioned as follows: drop each sample one by one due to vibration, move it to the camera section via a conveyor belt, determine and sort using the camera, and then the conveyor belt separates the acceptable products from the defective ones. **[Inspection Settings and Results]** We conducted inspections for shape defects using the samples we received. We verified using two patterns: from above and from the side, and it was possible to identify acceptable and defective products from above. The side inspection was able to detect the "angle of the edge," the "gap" in the clip section, and "twisting." The left image shows three types detected, while the right image displays the separated labels. There is no specified quantity for "OK," and the setting for defective is from 0 to 0 (if even one is present, it is considered NG). **[Software and Equipment Used]** Software used: DeepSky Learning Edition Field of view: Approximately 40 x 30 mm Minimum size of inspection target: 30 mm Number of inspection points: 3 Camera resolution: 1.3 million pixels Lens focal length: 25 mm Distance between lens and product: Approximately 170 mm Lighting: Indoor fluorescent lights

We will test the detection of scratches on metal sheets based on an inquiry from a copper manufacturer. This involves verification using provided photos. A copper sheet approximately 1mm thick is rolled up like toilet paper, and we are considering whether to inspect it while capturing images at low speed during the rolling process or to stop the line to take pictures and then inspect. In the first stage of a simple free evaluation, we were able to detect the scratched areas, but we are mistakenly detecting white areas on the image that are not scratches, making it difficult to distinguish. If we can improve the lighting to illuminate as wide an area as possible uniformly, inspection seems feasible. The images provided were used as "training data," and this is the result of the processing. Since these are training data images, we can make highly accurate judgments. The numbers represent the AI's confidence level percentage, referred to as "recognition points." [Software Used] Software: DeepSky Learning Version Number of inspection points: 1 across the entire screen

We received inquiries regarding multiple inspections, including unprocessed iron gear, pressure contact detection, burr residue on aluminum case parts, and confirmation of hole penetration in aluminum parts. Our company offers free evaluations that result in a simple "can do" or "cannot do" judgment. We can also accept paid testing. Please provide us with details about the specific operations. 【Inspection Settings and Results】 Continuous inspections are being conducted while manually rotating the workpiece slightly. Since the learning is not complete, there were some false detections, but the NG areas themselves were recognized. The ability to use the retraining function to improve the accuracy of inspection settings is one of the strengths of the software using AI (Deep Learning). 【Software Used】 Software used: DeepSky Learning Edition Number of inspection points: 1 (finding multiple defects from the entire screen)

Even with the same material and similar shapes, there are many cases where texture can be perceived and judged. We also sell to trading companies and industrial equipment manufacturers. Please feel free to contact us. 【Inspection Settings and Results】 It was possible to determine the presence or absence of processing (front or back) of the nuts through image processing using deep learning. Since there is almost no difference in the images between those that had chips removed and those that did not, they were treated as the same for inspection purposes, and a total of 20 pieces, 10 processed and 10 unprocessed, were used as training data, resulting in good judgment. Deep learning is one of the machine learning methods that teaches computers to learn the thinking processes that humans naturally perform. Our inspection software, DeepSky, is designed with this AI technology. 【Software Used】 Software Used: DeepSky Learning Version Number of Inspection Points: 1 (Determining whether the workpiece is front or back)

We received an inquiry about the inspection of fine precision machinery from a manufacturer engaged in the design of industrial machinery and the development of software. If sending sample products is difficult, it is also possible to verify using photographs. They were considering whether it could be automated to determine the sorting of good and defective products, which is currently done through visual inspection. 【Inspection Settings and Results】 It was possible to determine burrs and black spots on metal products. The report was based on a configuration using two cameras, each assumed to use a 0.5x macro lens and a 35mm prime lens. Even with verification through sent images, we are providing support as much as possible by listening to the details and suggesting imaging environments such as cameras. 【Software Used】 Software Used: DeepSky Learning Version Number of Inspection Points: One location across the entire screen (to find burrs and black spots)

We received an inspection request from a manufacturer of production equipment. The inspection is for determining the chipping of a metal product, specifically a "gear." They sent us images of 20 defective items and 5 acceptable items. For some workpieces that are difficult to send as samples, verification can sometimes be done through photographs. 【Inspection Settings and Results】 Using DeepSky's inspection function, we were able to accurately determine the chipping of the metal workpiece (gear). The process of setting up the areas to be detected by enclosing them in rectangles is called annotation; in this case, we only enclosed the defective parts for training. The images show the detection frames. The numbers indicate the AI's confidence level percentage (number of recognition points). If the number of recognition points is low or if there are misjudgments, further training can be conducted. DeepSky, released in 2020, is easy to set up, does not require fixed positioning, can detect various types of defects in shapes, and is particularly good at detecting defects in shiny workpieces like metal products. We have reported numerous evaluations of metal product inspections. 【Software Used】 Software Used: DeepSky Learning Version Number of Inspection Points: 1 (detecting defects from the entire screen)

We received a request for evaluation from a manufacturer of production equipment. This pertains to the judgment of defects caused by metal chips adhering to the workpiece. The DeepSky system released in 2020 is known for its ease of setup, the elimination of the need for fixed positioning, its ability to detect various shapes of defects, and its proficiency in detecting defects in shiny metal products, leading to numerous evaluations of metal product inspections. 【Inspection Setup and Results】 To determine the "metal chips inside the workpiece's hole," we made adjustments to the lighting. We attempted an inspection that captures images inside the hole, and this time, the bottom of the hole was clearly visible, allowing for accurate judgment. The initial evaluation is a report on whether simple detection is possible; however, depending on the position and shape of the defects, there may be instances where detection is challenging. After our free evaluation, we would like customers to directly experience the accuracy and setup methods before implementation, so we kindly ask you to utilize our free demo unit lending service. 【Software Used】 Software Used: DeepSky Learning Edition Minimum Size of Inspection Target: 2mm Number of Inspection Points: 1 (finding defects from the entire screen) Lighting: Coaxial Illumination

This is an inquiry from a manufacturer engaged in a wide variety of businesses, including iron gear processing, processing of irregular iron products, aluminum casting, aluminum processing, and assembly. I believe that many manufacturers, not just the customer in this inquiry, are still struggling with issues such as unprocessed materials, dents, pressure casting, and the detection of casting defects in aluminum cast products. Even if it is a work that was previously given up on, we welcome inquiries to our company. 【Inspection Settings and Results】 Using a software called DeepSky, which employs AI (Deep Learning), we were able to detect foreign objects in metal products. This software was released in 2020. By training the software on the areas we want to detect, it adjusts its own setting parameters and learns to recognize them. The image shows the detection frame, and the numbers indicate the AI's confidence level percentage and the "number of recognition points." 【Software Used】 Software used: DeepSky Learning Version Number of inspection points: 1 (to find foreign objects from the entire screen)

We received a request for inspection of cast precision parts from a manufacturer of production equipment design and fabrication. "Porosity" has been a frequently inquired case for some time. With DeepSky, released in 2020, it is easy to set up, does not require fixed positioning, can detect various defects with unstable shapes, and is also skilled at detecting defects in glossy workpieces. We have reported numerous evaluations of "porosity" inspections. [Inspection Setup and Results] This time, since the size of the defects was small, we narrowed the field of view to about 50mm, and DeepSky was able to recognize it well. Even if the patterns of defects to be detected increase, it seems likely that detection can still be achieved as long as the field of view is set appropriately. Our inspection software is also helpful for detecting fine defects in precision parts. [Software Used] Software Used: DeepSky Learning Edition Field of View: 50 x 40mm Minimum Size of Inspection Target: 0.5mm Number of Inspection Points: 1 (finding porosity from the entire screen)

It seems that manufacturers are struggling with various defect detections for saw blades, just like with other metal products, including issues such as unprocessed areas, dents, compression marks, chips, and foreign objects. For image inspection of shiny metal workpieces, please contact our company. This time, we are detecting wear on saw blades based on a verification request from a trading company we have been dealing with for some time. 【Inspection Settings and Results】 The judgment was made based on the NG images provided. Although we were unable to make sufficient settings due to difficulties in comparing with good products, we were still able to make some judgments, though some were misjudged. We report this as one result. The images show the defective areas that were successfully detected, highlighted with a light blue frame. The numbers indicate the confidence percentage (number of recognition points) from the inspection software. 【Software Used】 Software used: DeepSky Learning Edition Number of inspection points: 1 location (finding the worn area from the entire screen)

This is a verification to distinguish the defects of "scratches," "bleeding," and "spots" on metal painted products at the request of an industrial machinery manufacturer. We decided to photograph and verify by changing the orientation and position of a single defective sample. 【Inspection Settings and Results】 This time, we were able to make judgments by creatively adjusting the lighting during photography. Due to the size of the coaxial illumination we own, we could not capture the entire workpiece, so we focused on the damaged areas for imaging. Normally, inspections are conducted by capturing a field of view approximately 1.5 times the size with appropriately sized coaxial illumination, but we believe that the detection accuracy will be equivalent to that of this case based on the size of the damaged areas. To improve detection accuracy, we can use photos of misjudgments as training images for further learning. The left image is the "annotation" used to set parameters by enclosing the defective areas we want to detect. The right image is the "frame" detecting defects after learning and parameter setting.

We will verify whether we can detect "dents on the ball," "dents on the tapered section," and "indentations on the tapered section" based on a request from an industrial equipment manufacturer. By creating a striped pattern on the reflective part of the ball, it becomes easier to identify the dented areas due to the resulting step difference. 【Inspection Settings and Results】 We inspected 25 images, including 10 images of "dents on the ball," 5 images of "dents on the tapered section," 5 images of "indentations on the tapered section," and 5 images of "good products." Out of the 25 images, 22 were correctly identified as either good products or defective parts. The three images of "dents on the tapered section" could not be recognized as defective parts; however, considering that in actual operation, multiple shots (around 3 times) are expected to be taken during one full rotation, it does not necessarily mean that the defective parts that were not recognized in the verification will always go unrecognized.

In the manufacturing of metal press products such as automotive parts, there is often a process for attaching eyelets. This time, we will conduct inspections for the presence or absence of eyelets and rubber parts. We received good products and defective products that do not have all the parts attached. 【Inspection Settings and Results】 Using DeepSky's inspection function, we were able to classify good products and defective products without all the parts as negative by using them as teachers and marking random positions where eyelets are missing. It is set to pass only when the correct quantity of eyelets (and rubber parts) is detected. If even one eyelet is missing, it will be marked as negative due to quantity mismatch. If you want to check "which eyelet is missing," it is possible to use the area specification function, but for now, we conducted the inspection with settings to determine only OK or NG. The left image shows the work of enclosing the parts we want to teach, called "annotation." The right image is the detection frame image, where the numbers represent the AI's confidence level percentage, referred to as "recognition points."

Metal welded parts, such as automotive components, often have defects like bead misalignment, blowholes, and tears due to specification differences, which have been common inquiries for a long time. Previously, inspections were conducted using EasyInspector with fixed positioning, but with the inspection capabilities of DeepSky released in 2020, it has become easy to set up inspections without fixed positioning. 【Inspection Settings and Results】 The target areas to be detected were enclosed in frames and labeled by type. A total of 14 images were used as training data, consisting of 8 OK images and 6 NG images. The learning process was executed for 2,000 steps, and the graph converged in about 16 minutes. The time may vary depending on the specifications of the PC used. In this inspection, since we are detecting the defective areas that were trained, it was set so that if even one defect is detected, it would be considered NG (only OK when the count is from 0 to 0), allowing for the detection of welding defects.

In the metal product manufacturing industry, deformation caused by spatter has been a challenge for some time. Recently, there was an incident where defective products were generated due to spatter occurring on the contact surface, leading to inquiries through a trading company. This customer has been using our inspection software for some time. 【Inspection Settings and Results】 We were able to detect the issues using software called DeepSky, which utilizes AI. DeepSky excels at inspections that involve searching for defects throughout the entire screen. The image on the left is a teacher image that trains the AI on the defective areas, while the right side shows the detected image, with the numbers indicating the AI's confidence level in percentage (number of recognition points). Since the spatter defects were very small and it was difficult to inspect the entire workpiece with a single camera, we are proposing inspections using multiple cameras. 【Software and Equipment Used】 Software Used: DeepSky Field of View: 70 x 60 mm Minimum Size of Inspection Target: 1 mm Number of Inspection Points: 1 overall Camera Resolution: 1.3 million pixels Lens Focal Length: 6 mm Distance between Lens and Product: Approximately 110 mm Lighting: Ring lighting Distance from Lighting to Inspection Item: Approximately 300 mm from above

Even manufacturers who specialize in precision cutting and grinding processes are utilizing our inspection software for high-quality precision machining technology and quality management systems. 【Inspection Settings and Results】 As a result of verification using the samples provided, it was possible to detect scratches on the screws. The inspection was conducted using a software called DeepSky, which employs AI (Deep Learning). By training the software to recognize the desired scratches, it adjusts its own setting parameters to identify them. Five samples were photographed from different angles and the opposite side, resulting in 22 training images, with the left image set as the reference. The right image shows the detection frame. 【Software and Equipment Used】 Software Used: DeepSky Field of View: 30 x 25mm Minimum Size of Inspection Target: 0.2mm Number of Inspection Points: 1 overall Camera Resolution: 1.3 million pixels Lens Focal Length: 35mm + 5mm close-up ring Distance Between Lens and Product: Approximately 160mm Lighting: Indoor fluorescent lights

"Marks from screws" is a very common request for verification. While it may be better to inspect using our conventional general-purpose image inspection software "EasyInspector," in operations where judgments are made on a conveyor or during work, we sometimes recommend using the AI-based inspection software "DeepSky" for defects like marks that come in various colors and shapes. This time, we created a thread conveyor and reported on the detection of defects around the screws in a full 360-degree rotation. [Inspection Settings and Results] The task of enclosing the area we want to detect (in this case, marks) in a rectangle is called annotation. We trained the model with 24 annotated teacher images, rotating them approximately 2300 steps (about 15 minutes). The training time varies depending on the specifications of the PC. In the right image, the marked area is indicated by a green detection frame.

Manufacturers of monitoring and measurement devices are considering our inspection software. We conducted a free evaluation for detecting scratches on the device's hardware. Sending actual samples allows us to provide the most reliable report. In that case, you will need to send both good products and the workpieces with defects you want to detect. Please contact us for more details. 【Inspection Settings and Results】 The process of enclosing the areas to be detected in a rectangle, as shown in the left image, is called annotation. Since the scratches were of different types, we labeled and registered them as "Scratch A / Scratch B / Scratch C." Although it was not part of the requested inspection items, we were able to capture a small dent on the cylindrical part clearly, so it was included as a detection target (Scratch C). (Detection will not occur unless annotated.) The clip was fixed, so the clip area was set outside the field of view. 【Software and Equipment Used】 Software Used: DeepSky Learning Edition Field of View: Approximately 64 x 51 mm Minimum Size of Inspection Target: 20 mm Number of Inspection Points: 1 in total Camera Resolution: 1.3 million pixels Lens Focal Length: 25 mm Distance Between Lens and Product: Approximately 260 mm Lighting: Ring Lighting

Deformation of the insertion port in cassette gas products can lead to significant accidents. We hope that our inspection software will be useful for your safety. In this free evaluation, we used the software DeepSky, which employs AI (Deep Learning) for inspection. By training the software on the areas we want to detect, it adjusts its own setting parameters and learns to recognize them. [Inspection Settings and Results] By using DeepSky's inspection capabilities, we determined multiple types of defects. We assessed the presence or absence of each type of defect based on scratches, shadows, and reflections, achieving a judgment time of 0.33 seconds. The labels were divided into three categories: scratches, shadows, and reflections, and we set up the software to learn the specific defects we wanted to identify. The method of specifying these defects and how to capture the images is crucial for image inspection.

Our inspection software users include manufacturers of precision motors. They focus on precision casting, emphasizing technology and experience. We are pleased that our inspection software can contribute to such high-performance, precise manufacturing. [Inspection Settings and Results] Settings were made with DeepSky. When illuminated, the area closest to the camera in the center glows white; this is registered as OK, and when OK is detected, it is considered a pass. Even in areas where OK cannot be detected, if NG is not detected, it is set to pass, allowing us to meet the customer's judgment requirements. The image shows the types of labels (for distinguishing the parts to be detected). [Software and Equipment Used] Software Used: DeepSky 2.0 Field of View: Approximately 48 x 36 mm Minimum Size of Inspection Target: 20 mm Number of Inspection Points: 1 (Label 4) Camera Resolution: 1.3 million pixels Lens Focal Length: 12 mm Distance Between Lens and Product: 130 mm Lighting: Ring Lighting Distance Between Lighting and Product: 130 mm

To ensure stable quality during shipment, many of our customers operate at low cost using our inspection software. 【Inspection Settings and Results】 By using the "Comparison with Master Image" feature of EasyInspector, we detected differences in two areas: "unwelded" and "punctured." Additionally, using the "Scratch and Defect Inspection" feature, we identified the black, band-like charred areas that occurred during "sagging." The inspection cycle time was 0.25 seconds. 【Software and Equipment Used】 Software Used: EasyInspector310 Field of View: Approximately 13 x 10mm Minimum Size of Inspection Target: 1mm Number of Inspection Points: 3 Camera Resolution: 1.3 Megapixels Lens Focal Length: 35mm + 10mm Macro Ring Distance Between Lens and Product: 130mm Lighting: Indoor Fluorescent Light The current 'EasyInspector2' color package allows inspection with the features of [Comparison with Master Image] and [Scratch and Defect Detection].

This is a request for a free evaluation of cast products. There was an inquiry about wanting to incorporate inspections if they can be done at a low cost. There are still many inquiries about die-cast and cast products; nowadays, we often suggest easy settings using the AI-based inspection software DeepSky. If you are inspecting workpieces made of reflective materials like cast products while flowing them on a conveyor, please check "DeepSky." 【Inspection Settings and Results】 By using the "Scratch Inspection" feature of EasyInspector, we were able to detect the presence or absence of defects at one location (simultaneous inspection of three workpieces) and determine visually similar items (different products) in 0.3 seconds. The green numbers indicate OK (good products) within the acceptable range, while the red text indicates NG (defective). The settings for good and defective products can be changed.

We received an inquiry from a thermosensor manufacturing company regarding the automation of shape verification. This is a free evaluation based on the images you provided. Due to the spread of the COVID-19 virus, products that detect temperature are increasing in variety. We encounter new forms of the latest surface temperature detection systems in various locations. This free evaluation is from 2017, but it reaffirms that our inspection software is being utilized in various scenarios. 【Inspection Settings and Results】 By using the "Comparison with Master Image" feature of EasyInspector, we were able to determine visually similar items (different products) in less than 0.48 seconds. In actual inspections, the appearance may change due to factors such as lighting reflections, which could affect the inspection results. For inspection items labeled "unprocessed" or "punctured," the shapes differ, so inspection methods using backlighting may be required. Additionally, in the inspection of "sagging," the size of the sagging may affect the inspection results.

In recent years, system security manufacturers have been increasingly expanding and are also engaged in the production of hardware products. This involves inspecting the presence or absence of bit inserts in covers that encase the equipment. 【Inspection Settings and Results】 By using the "Color Comparison Inspection" feature of EasyInspector, we were able to detect the differences in the presence or absence of bit inserts in four holes and determine visually similar counterfeit products (different items) in just 0.25 seconds. Our inspection software may also be of assistance on your production line. We look forward to your inquiries. 【Software and Equipment Used】 Software Used: EasyInspector710 Field of View: 367 x 292 mm Minimum Size of Inspection Target: 10 mm Number of Inspection Points: 4 Camera Resolution: 1.3 Megapixels Lens Focal Length: 6 mm Distance from Lens to Product: 330 mm Lighting: Ring Lighting Distance from Lighting to Inspection Item: Approximately 280 mm Current inspections can be performed with the 'EasyInspector2' color package [Color Comparison Extraction + Particle Count].

This is an inquiry regarding the inspection of die-cast products during conveyor transport, with specific parameters: hole diameter tolerance of ±0.05, overall tolerance of ±0.4, inspection cycle time of under 10 seconds, and a minimum conveyor speed of 1,000 mm/min. The inspection focused on the presence or absence of burrs, chips, and holes. Die-cast products have been a frequently asked topic at our company recently, and there were many inquiries even as far back as 2017. At that time, projects that could now be easily and accurately inspected using AI-based inspection software DeepSky were still being inspected with the conventional inspection software EasyInspector. [Inspection Settings and Results] By using the "Comparison with Master Image" function of EasyInspector, we proposed setting inspection frames at two locations within the camera's field of view, utilizing multiple cameras for inspection. The cycle time for a single capture and inspection was approximately 0.54 seconds.

This inquiry is about molded metal parts, but producing products with very similar shapes in the same factory is common across all industries. To distinguish between finely similar parts with differences of less than 1mm or to identify items that look exactly the same from above, we recommend using our inspection software, EasyInspector. 【Inspection Settings and Results】 By using the "Dimension and Angle Inspection" feature of EasyInspector, we were able to detect a slight dimensional difference at one location and determine visually similar items (different products) in 0.17 seconds. In the image above, the red line represents the inspection frame, while the green line indicates the area being measured. Measurements are taken between the green lines. 【Software and Equipment Used】 Software Used: EasyInspector710 Field of View: 32x 20mm Minimum Size of Inspection Target: 0.5mm Number of Inspection Points: 1 Camera Resolution: 3 million pixels Lens Focal Length: 50mm + 10mm close-up ring Distance Between Lens and Product: 230mm Lighting: Backlight Current 'EasyInspector2' MS (MeaSure) package can perform inspections with [Position and Width Measurement] and [Angle Measurement].

It is common in various industries to manufacture products with multiple models on the same line. This time, we inspected whether similar parts were incorrectly installed. The customer, a meter manufacturing company that contacted us, was having trouble with the installation of similar parts. 【Inspection Settings and Results】 By using EasyInspector's "Presence of Specified Color Inspection" feature, we were able to detect a single difference and determine visually similar counterfeit products (different items) in less than 0.13 seconds. A key point for this inspection was positioning the camera diagonally above the inspection item. 【Software and Equipment Used】 Software Used: EasyInspector710 Field of View: 400 x 300mm Minimum Size of Inspection Target: 2mm Number of Inspection Points: 1 Camera Resolution: 300,000 pixels Lens Focal Length: 25mm Distance Between Lens and Product: 320mm Lighting: Fluorescent lighting, approximately 300mm distance to the inspection area The current 'EasyInspector2' color package can perform inspections for the "Presence of Specified Color."

The metal parts used by automobile manufacturers may sometimes be shipped with defects such as dents or black spots, but if the dimensions are incorrect, it is pointless. This led us to use our software to inspect those dimensions. 【Inspection Settings and Results】 Using the "Dimension Angle Inspection" feature of EasyInspector, we measured three different areas with varying lengths. First, to clearly define the edges, we used backlighting; however, when the inspection object was placed directly on the backlight, the light reflected irregularly, causing the edges to appear blurred. Therefore, we took the photos at a distance of about 40mm from the backlight, placing the object on a glass surface. Since the inspection object appeared dark, we detected the bright and dark areas from top to bottom to measure the dimensions.